xref: /openbmc/linux/arch/powerpc/kernel/traps.c (revision 8795a739)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  Copyright (C) 1995-1996  Gary Thomas (gdt@linuxppc.org)
4  *  Copyright 2007-2010 Freescale Semiconductor, Inc.
5  *
6  *  Modified by Cort Dougan (cort@cs.nmt.edu)
7  *  and Paul Mackerras (paulus@samba.org)
8  */
9 
10 /*
11  * This file handles the architecture-dependent parts of hardware exceptions
12  */
13 
14 #include <linux/errno.h>
15 #include <linux/sched.h>
16 #include <linux/sched/debug.h>
17 #include <linux/kernel.h>
18 #include <linux/mm.h>
19 #include <linux/pkeys.h>
20 #include <linux/stddef.h>
21 #include <linux/unistd.h>
22 #include <linux/ptrace.h>
23 #include <linux/user.h>
24 #include <linux/interrupt.h>
25 #include <linux/init.h>
26 #include <linux/extable.h>
27 #include <linux/module.h>	/* print_modules */
28 #include <linux/prctl.h>
29 #include <linux/delay.h>
30 #include <linux/kprobes.h>
31 #include <linux/kexec.h>
32 #include <linux/backlight.h>
33 #include <linux/bug.h>
34 #include <linux/kdebug.h>
35 #include <linux/ratelimit.h>
36 #include <linux/context_tracking.h>
37 #include <linux/smp.h>
38 #include <linux/console.h>
39 #include <linux/kmsg_dump.h>
40 
41 #include <asm/emulated_ops.h>
42 #include <asm/pgtable.h>
43 #include <linux/uaccess.h>
44 #include <asm/debugfs.h>
45 #include <asm/io.h>
46 #include <asm/machdep.h>
47 #include <asm/rtas.h>
48 #include <asm/pmc.h>
49 #include <asm/reg.h>
50 #ifdef CONFIG_PMAC_BACKLIGHT
51 #include <asm/backlight.h>
52 #endif
53 #ifdef CONFIG_PPC64
54 #include <asm/firmware.h>
55 #include <asm/processor.h>
56 #include <asm/tm.h>
57 #endif
58 #include <asm/kexec.h>
59 #include <asm/ppc-opcode.h>
60 #include <asm/rio.h>
61 #include <asm/fadump.h>
62 #include <asm/switch_to.h>
63 #include <asm/tm.h>
64 #include <asm/debug.h>
65 #include <asm/asm-prototypes.h>
66 #include <asm/hmi.h>
67 #include <sysdev/fsl_pci.h>
68 #include <asm/kprobes.h>
69 #include <asm/stacktrace.h>
70 #include <asm/nmi.h>
71 
72 #if defined(CONFIG_DEBUGGER) || defined(CONFIG_KEXEC_CORE)
73 int (*__debugger)(struct pt_regs *regs) __read_mostly;
74 int (*__debugger_ipi)(struct pt_regs *regs) __read_mostly;
75 int (*__debugger_bpt)(struct pt_regs *regs) __read_mostly;
76 int (*__debugger_sstep)(struct pt_regs *regs) __read_mostly;
77 int (*__debugger_iabr_match)(struct pt_regs *regs) __read_mostly;
78 int (*__debugger_break_match)(struct pt_regs *regs) __read_mostly;
79 int (*__debugger_fault_handler)(struct pt_regs *regs) __read_mostly;
80 
81 EXPORT_SYMBOL(__debugger);
82 EXPORT_SYMBOL(__debugger_ipi);
83 EXPORT_SYMBOL(__debugger_bpt);
84 EXPORT_SYMBOL(__debugger_sstep);
85 EXPORT_SYMBOL(__debugger_iabr_match);
86 EXPORT_SYMBOL(__debugger_break_match);
87 EXPORT_SYMBOL(__debugger_fault_handler);
88 #endif
89 
90 /* Transactional Memory trap debug */
91 #ifdef TM_DEBUG_SW
92 #define TM_DEBUG(x...) printk(KERN_INFO x)
93 #else
94 #define TM_DEBUG(x...) do { } while(0)
95 #endif
96 
97 static const char *signame(int signr)
98 {
99 	switch (signr) {
100 	case SIGBUS:	return "bus error";
101 	case SIGFPE:	return "floating point exception";
102 	case SIGILL:	return "illegal instruction";
103 	case SIGSEGV:	return "segfault";
104 	case SIGTRAP:	return "unhandled trap";
105 	}
106 
107 	return "unknown signal";
108 }
109 
110 /*
111  * Trap & Exception support
112  */
113 
114 #ifdef CONFIG_PMAC_BACKLIGHT
115 static void pmac_backlight_unblank(void)
116 {
117 	mutex_lock(&pmac_backlight_mutex);
118 	if (pmac_backlight) {
119 		struct backlight_properties *props;
120 
121 		props = &pmac_backlight->props;
122 		props->brightness = props->max_brightness;
123 		props->power = FB_BLANK_UNBLANK;
124 		backlight_update_status(pmac_backlight);
125 	}
126 	mutex_unlock(&pmac_backlight_mutex);
127 }
128 #else
129 static inline void pmac_backlight_unblank(void) { }
130 #endif
131 
132 /*
133  * If oops/die is expected to crash the machine, return true here.
134  *
135  * This should not be expected to be 100% accurate, there may be
136  * notifiers registered or other unexpected conditions that may bring
137  * down the kernel. Or if the current process in the kernel is holding
138  * locks or has other critical state, the kernel may become effectively
139  * unusable anyway.
140  */
141 bool die_will_crash(void)
142 {
143 	if (should_fadump_crash())
144 		return true;
145 	if (kexec_should_crash(current))
146 		return true;
147 	if (in_interrupt() || panic_on_oops ||
148 			!current->pid || is_global_init(current))
149 		return true;
150 
151 	return false;
152 }
153 
154 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
155 static int die_owner = -1;
156 static unsigned int die_nest_count;
157 static int die_counter;
158 
159 extern void panic_flush_kmsg_start(void)
160 {
161 	/*
162 	 * These are mostly taken from kernel/panic.c, but tries to do
163 	 * relatively minimal work. Don't use delay functions (TB may
164 	 * be broken), don't crash dump (need to set a firmware log),
165 	 * don't run notifiers. We do want to get some information to
166 	 * Linux console.
167 	 */
168 	console_verbose();
169 	bust_spinlocks(1);
170 }
171 
172 extern void panic_flush_kmsg_end(void)
173 {
174 	printk_safe_flush_on_panic();
175 	kmsg_dump(KMSG_DUMP_PANIC);
176 	bust_spinlocks(0);
177 	debug_locks_off();
178 	console_flush_on_panic(CONSOLE_FLUSH_PENDING);
179 }
180 
181 static unsigned long oops_begin(struct pt_regs *regs)
182 {
183 	int cpu;
184 	unsigned long flags;
185 
186 	oops_enter();
187 
188 	/* racy, but better than risking deadlock. */
189 	raw_local_irq_save(flags);
190 	cpu = smp_processor_id();
191 	if (!arch_spin_trylock(&die_lock)) {
192 		if (cpu == die_owner)
193 			/* nested oops. should stop eventually */;
194 		else
195 			arch_spin_lock(&die_lock);
196 	}
197 	die_nest_count++;
198 	die_owner = cpu;
199 	console_verbose();
200 	bust_spinlocks(1);
201 	if (machine_is(powermac))
202 		pmac_backlight_unblank();
203 	return flags;
204 }
205 NOKPROBE_SYMBOL(oops_begin);
206 
207 static void oops_end(unsigned long flags, struct pt_regs *regs,
208 			       int signr)
209 {
210 	bust_spinlocks(0);
211 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
212 	die_nest_count--;
213 	oops_exit();
214 	printk("\n");
215 	if (!die_nest_count) {
216 		/* Nest count reaches zero, release the lock. */
217 		die_owner = -1;
218 		arch_spin_unlock(&die_lock);
219 	}
220 	raw_local_irq_restore(flags);
221 
222 	/*
223 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
224 	 */
225 	if (TRAP(regs) == 0x100)
226 		return;
227 
228 	crash_fadump(regs, "die oops");
229 
230 	if (kexec_should_crash(current))
231 		crash_kexec(regs);
232 
233 	if (!signr)
234 		return;
235 
236 	/*
237 	 * While our oops output is serialised by a spinlock, output
238 	 * from panic() called below can race and corrupt it. If we
239 	 * know we are going to panic, delay for 1 second so we have a
240 	 * chance to get clean backtraces from all CPUs that are oopsing.
241 	 */
242 	if (in_interrupt() || panic_on_oops || !current->pid ||
243 	    is_global_init(current)) {
244 		mdelay(MSEC_PER_SEC);
245 	}
246 
247 	if (panic_on_oops)
248 		panic("Fatal exception");
249 	do_exit(signr);
250 }
251 NOKPROBE_SYMBOL(oops_end);
252 
253 static char *get_mmu_str(void)
254 {
255 	if (early_radix_enabled())
256 		return " MMU=Radix";
257 	if (early_mmu_has_feature(MMU_FTR_HPTE_TABLE))
258 		return " MMU=Hash";
259 	return "";
260 }
261 
262 static int __die(const char *str, struct pt_regs *regs, long err)
263 {
264 	printk("Oops: %s, sig: %ld [#%d]\n", str, err, ++die_counter);
265 
266 	printk("%s PAGE_SIZE=%luK%s%s%s%s%s%s %s\n",
267 	       IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN) ? "LE" : "BE",
268 	       PAGE_SIZE / 1024, get_mmu_str(),
269 	       IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT" : "",
270 	       IS_ENABLED(CONFIG_SMP) ? " SMP" : "",
271 	       IS_ENABLED(CONFIG_SMP) ? (" NR_CPUS=" __stringify(NR_CPUS)) : "",
272 	       debug_pagealloc_enabled() ? " DEBUG_PAGEALLOC" : "",
273 	       IS_ENABLED(CONFIG_NUMA) ? " NUMA" : "",
274 	       ppc_md.name ? ppc_md.name : "");
275 
276 	if (notify_die(DIE_OOPS, str, regs, err, 255, SIGSEGV) == NOTIFY_STOP)
277 		return 1;
278 
279 	print_modules();
280 	show_regs(regs);
281 
282 	return 0;
283 }
284 NOKPROBE_SYMBOL(__die);
285 
286 void die(const char *str, struct pt_regs *regs, long err)
287 {
288 	unsigned long flags;
289 
290 	/*
291 	 * system_reset_excption handles debugger, crash dump, panic, for 0x100
292 	 */
293 	if (TRAP(regs) != 0x100) {
294 		if (debugger(regs))
295 			return;
296 	}
297 
298 	flags = oops_begin(regs);
299 	if (__die(str, regs, err))
300 		err = 0;
301 	oops_end(flags, regs, err);
302 }
303 NOKPROBE_SYMBOL(die);
304 
305 void user_single_step_report(struct pt_regs *regs)
306 {
307 	force_sig_fault(SIGTRAP, TRAP_TRACE, (void __user *)regs->nip);
308 }
309 
310 static void show_signal_msg(int signr, struct pt_regs *regs, int code,
311 			    unsigned long addr)
312 {
313 	static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
314 				      DEFAULT_RATELIMIT_BURST);
315 
316 	if (!show_unhandled_signals)
317 		return;
318 
319 	if (!unhandled_signal(current, signr))
320 		return;
321 
322 	if (!__ratelimit(&rs))
323 		return;
324 
325 	pr_info("%s[%d]: %s (%d) at %lx nip %lx lr %lx code %x",
326 		current->comm, current->pid, signame(signr), signr,
327 		addr, regs->nip, regs->link, code);
328 
329 	print_vma_addr(KERN_CONT " in ", regs->nip);
330 
331 	pr_cont("\n");
332 
333 	show_user_instructions(regs);
334 }
335 
336 static bool exception_common(int signr, struct pt_regs *regs, int code,
337 			      unsigned long addr)
338 {
339 	if (!user_mode(regs)) {
340 		die("Exception in kernel mode", regs, signr);
341 		return false;
342 	}
343 
344 	show_signal_msg(signr, regs, code, addr);
345 
346 	if (arch_irqs_disabled() && !arch_irq_disabled_regs(regs))
347 		local_irq_enable();
348 
349 	current->thread.trap_nr = code;
350 
351 	/*
352 	 * Save all the pkey registers AMR/IAMR/UAMOR. Eg: Core dumps need
353 	 * to capture the content, if the task gets killed.
354 	 */
355 	thread_pkey_regs_save(&current->thread);
356 
357 	return true;
358 }
359 
360 void _exception_pkey(struct pt_regs *regs, unsigned long addr, int key)
361 {
362 	if (!exception_common(SIGSEGV, regs, SEGV_PKUERR, addr))
363 		return;
364 
365 	force_sig_pkuerr((void __user *) addr, key);
366 }
367 
368 void _exception(int signr, struct pt_regs *regs, int code, unsigned long addr)
369 {
370 	if (!exception_common(signr, regs, code, addr))
371 		return;
372 
373 	force_sig_fault(signr, code, (void __user *)addr);
374 }
375 
376 /*
377  * The interrupt architecture has a quirk in that the HV interrupts excluding
378  * the NMIs (0x100 and 0x200) do not clear MSR[RI] at entry. The first thing
379  * that an interrupt handler must do is save off a GPR into a scratch register,
380  * and all interrupts on POWERNV (HV=1) use the HSPRG1 register as scratch.
381  * Therefore an NMI can clobber an HV interrupt's live HSPRG1 without noticing
382  * that it is non-reentrant, which leads to random data corruption.
383  *
384  * The solution is for NMI interrupts in HV mode to check if they originated
385  * from these critical HV interrupt regions. If so, then mark them not
386  * recoverable.
387  *
388  * An alternative would be for HV NMIs to use SPRG for scratch to avoid the
389  * HSPRG1 clobber, however this would cause guest SPRG to be clobbered. Linux
390  * guests should always have MSR[RI]=0 when its scratch SPRG is in use, so
391  * that would work. However any other guest OS that may have the SPRG live
392  * and MSR[RI]=1 could encounter silent corruption.
393  *
394  * Builds that do not support KVM could take this second option to increase
395  * the recoverability of NMIs.
396  */
397 void hv_nmi_check_nonrecoverable(struct pt_regs *regs)
398 {
399 #ifdef CONFIG_PPC_POWERNV
400 	unsigned long kbase = (unsigned long)_stext;
401 	unsigned long nip = regs->nip;
402 
403 	if (!(regs->msr & MSR_RI))
404 		return;
405 	if (!(regs->msr & MSR_HV))
406 		return;
407 	if (regs->msr & MSR_PR)
408 		return;
409 
410 	/*
411 	 * Now test if the interrupt has hit a range that may be using
412 	 * HSPRG1 without having RI=0 (i.e., an HSRR interrupt). The
413 	 * problem ranges all run un-relocated. Test real and virt modes
414 	 * at the same time by droping the high bit of the nip (virt mode
415 	 * entry points still have the +0x4000 offset).
416 	 */
417 	nip &= ~0xc000000000000000ULL;
418 	if ((nip >= 0x500 && nip < 0x600) || (nip >= 0x4500 && nip < 0x4600))
419 		goto nonrecoverable;
420 	if ((nip >= 0x980 && nip < 0xa00) || (nip >= 0x4980 && nip < 0x4a00))
421 		goto nonrecoverable;
422 	if ((nip >= 0xe00 && nip < 0xec0) || (nip >= 0x4e00 && nip < 0x4ec0))
423 		goto nonrecoverable;
424 	if ((nip >= 0xf80 && nip < 0xfa0) || (nip >= 0x4f80 && nip < 0x4fa0))
425 		goto nonrecoverable;
426 
427 	/* Trampoline code runs un-relocated so subtract kbase. */
428 	if (nip >= (unsigned long)(start_real_trampolines - kbase) &&
429 			nip < (unsigned long)(end_real_trampolines - kbase))
430 		goto nonrecoverable;
431 	if (nip >= (unsigned long)(start_virt_trampolines - kbase) &&
432 			nip < (unsigned long)(end_virt_trampolines - kbase))
433 		goto nonrecoverable;
434 	return;
435 
436 nonrecoverable:
437 	regs->msr &= ~MSR_RI;
438 #endif
439 }
440 
441 void system_reset_exception(struct pt_regs *regs)
442 {
443 	unsigned long hsrr0, hsrr1;
444 	bool nested = in_nmi();
445 	bool saved_hsrrs = false;
446 
447 	/*
448 	 * Avoid crashes in case of nested NMI exceptions. Recoverability
449 	 * is determined by RI and in_nmi
450 	 */
451 	if (!nested)
452 		nmi_enter();
453 
454 	/*
455 	 * System reset can interrupt code where HSRRs are live and MSR[RI]=1.
456 	 * The system reset interrupt itself may clobber HSRRs (e.g., to call
457 	 * OPAL), so save them here and restore them before returning.
458 	 *
459 	 * Machine checks don't need to save HSRRs, as the real mode handler
460 	 * is careful to avoid them, and the regular handler is not delivered
461 	 * as an NMI.
462 	 */
463 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
464 		hsrr0 = mfspr(SPRN_HSRR0);
465 		hsrr1 = mfspr(SPRN_HSRR1);
466 		saved_hsrrs = true;
467 	}
468 
469 	hv_nmi_check_nonrecoverable(regs);
470 
471 	__this_cpu_inc(irq_stat.sreset_irqs);
472 
473 	/* See if any machine dependent calls */
474 	if (ppc_md.system_reset_exception) {
475 		if (ppc_md.system_reset_exception(regs))
476 			goto out;
477 	}
478 
479 	if (debugger(regs))
480 		goto out;
481 
482 	kmsg_dump(KMSG_DUMP_OOPS);
483 	/*
484 	 * A system reset is a request to dump, so we always send
485 	 * it through the crashdump code (if fadump or kdump are
486 	 * registered).
487 	 */
488 	crash_fadump(regs, "System Reset");
489 
490 	crash_kexec(regs);
491 
492 	/*
493 	 * We aren't the primary crash CPU. We need to send it
494 	 * to a holding pattern to avoid it ending up in the panic
495 	 * code.
496 	 */
497 	crash_kexec_secondary(regs);
498 
499 	/*
500 	 * No debugger or crash dump registered, print logs then
501 	 * panic.
502 	 */
503 	die("System Reset", regs, SIGABRT);
504 
505 	mdelay(2*MSEC_PER_SEC); /* Wait a little while for others to print */
506 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
507 	nmi_panic(regs, "System Reset");
508 
509 out:
510 #ifdef CONFIG_PPC_BOOK3S_64
511 	BUG_ON(get_paca()->in_nmi == 0);
512 	if (get_paca()->in_nmi > 1)
513 		nmi_panic(regs, "Unrecoverable nested System Reset");
514 #endif
515 	/* Must die if the interrupt is not recoverable */
516 	if (!(regs->msr & MSR_RI))
517 		nmi_panic(regs, "Unrecoverable System Reset");
518 
519 	if (saved_hsrrs) {
520 		mtspr(SPRN_HSRR0, hsrr0);
521 		mtspr(SPRN_HSRR1, hsrr1);
522 	}
523 
524 	if (!nested)
525 		nmi_exit();
526 
527 	/* What should we do here? We could issue a shutdown or hard reset. */
528 }
529 
530 /*
531  * I/O accesses can cause machine checks on powermacs.
532  * Check if the NIP corresponds to the address of a sync
533  * instruction for which there is an entry in the exception
534  * table.
535  * Note that the 601 only takes a machine check on TEA
536  * (transfer error ack) signal assertion, and does not
537  * set any of the top 16 bits of SRR1.
538  *  -- paulus.
539  */
540 static inline int check_io_access(struct pt_regs *regs)
541 {
542 #ifdef CONFIG_PPC32
543 	unsigned long msr = regs->msr;
544 	const struct exception_table_entry *entry;
545 	unsigned int *nip = (unsigned int *)regs->nip;
546 
547 	if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
548 	    && (entry = search_exception_tables(regs->nip)) != NULL) {
549 		/*
550 		 * Check that it's a sync instruction, or somewhere
551 		 * in the twi; isync; nop sequence that inb/inw/inl uses.
552 		 * As the address is in the exception table
553 		 * we should be able to read the instr there.
554 		 * For the debug message, we look at the preceding
555 		 * load or store.
556 		 */
557 		if (*nip == PPC_INST_NOP)
558 			nip -= 2;
559 		else if (*nip == PPC_INST_ISYNC)
560 			--nip;
561 		if (*nip == PPC_INST_SYNC || (*nip >> 26) == OP_TRAP) {
562 			unsigned int rb;
563 
564 			--nip;
565 			rb = (*nip >> 11) & 0x1f;
566 			printk(KERN_DEBUG "%s bad port %lx at %p\n",
567 			       (*nip & 0x100)? "OUT to": "IN from",
568 			       regs->gpr[rb] - _IO_BASE, nip);
569 			regs->msr |= MSR_RI;
570 			regs->nip = extable_fixup(entry);
571 			return 1;
572 		}
573 	}
574 #endif /* CONFIG_PPC32 */
575 	return 0;
576 }
577 
578 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
579 /* On 4xx, the reason for the machine check or program exception
580    is in the ESR. */
581 #define get_reason(regs)	((regs)->dsisr)
582 #define REASON_FP		ESR_FP
583 #define REASON_ILLEGAL		(ESR_PIL | ESR_PUO)
584 #define REASON_PRIVILEGED	ESR_PPR
585 #define REASON_TRAP		ESR_PTR
586 
587 /* single-step stuff */
588 #define single_stepping(regs)	(current->thread.debug.dbcr0 & DBCR0_IC)
589 #define clear_single_step(regs)	(current->thread.debug.dbcr0 &= ~DBCR0_IC)
590 #define clear_br_trace(regs)	do {} while(0)
591 #else
592 /* On non-4xx, the reason for the machine check or program
593    exception is in the MSR. */
594 #define get_reason(regs)	((regs)->msr)
595 #define REASON_TM		SRR1_PROGTM
596 #define REASON_FP		SRR1_PROGFPE
597 #define REASON_ILLEGAL		SRR1_PROGILL
598 #define REASON_PRIVILEGED	SRR1_PROGPRIV
599 #define REASON_TRAP		SRR1_PROGTRAP
600 
601 #define single_stepping(regs)	((regs)->msr & MSR_SE)
602 #define clear_single_step(regs)	((regs)->msr &= ~MSR_SE)
603 #define clear_br_trace(regs)	((regs)->msr &= ~MSR_BE)
604 #endif
605 
606 #if defined(CONFIG_E500)
607 int machine_check_e500mc(struct pt_regs *regs)
608 {
609 	unsigned long mcsr = mfspr(SPRN_MCSR);
610 	unsigned long pvr = mfspr(SPRN_PVR);
611 	unsigned long reason = mcsr;
612 	int recoverable = 1;
613 
614 	if (reason & MCSR_LD) {
615 		recoverable = fsl_rio_mcheck_exception(regs);
616 		if (recoverable == 1)
617 			goto silent_out;
618 	}
619 
620 	printk("Machine check in kernel mode.\n");
621 	printk("Caused by (from MCSR=%lx): ", reason);
622 
623 	if (reason & MCSR_MCP)
624 		pr_cont("Machine Check Signal\n");
625 
626 	if (reason & MCSR_ICPERR) {
627 		pr_cont("Instruction Cache Parity Error\n");
628 
629 		/*
630 		 * This is recoverable by invalidating the i-cache.
631 		 */
632 		mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI);
633 		while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI)
634 			;
635 
636 		/*
637 		 * This will generally be accompanied by an instruction
638 		 * fetch error report -- only treat MCSR_IF as fatal
639 		 * if it wasn't due to an L1 parity error.
640 		 */
641 		reason &= ~MCSR_IF;
642 	}
643 
644 	if (reason & MCSR_DCPERR_MC) {
645 		pr_cont("Data Cache Parity Error\n");
646 
647 		/*
648 		 * In write shadow mode we auto-recover from the error, but it
649 		 * may still get logged and cause a machine check.  We should
650 		 * only treat the non-write shadow case as non-recoverable.
651 		 */
652 		/* On e6500 core, L1 DCWS (Data cache write shadow mode) bit
653 		 * is not implemented but L1 data cache always runs in write
654 		 * shadow mode. Hence on data cache parity errors HW will
655 		 * automatically invalidate the L1 Data Cache.
656 		 */
657 		if (PVR_VER(pvr) != PVR_VER_E6500) {
658 			if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS))
659 				recoverable = 0;
660 		}
661 	}
662 
663 	if (reason & MCSR_L2MMU_MHIT) {
664 		pr_cont("Hit on multiple TLB entries\n");
665 		recoverable = 0;
666 	}
667 
668 	if (reason & MCSR_NMI)
669 		pr_cont("Non-maskable interrupt\n");
670 
671 	if (reason & MCSR_IF) {
672 		pr_cont("Instruction Fetch Error Report\n");
673 		recoverable = 0;
674 	}
675 
676 	if (reason & MCSR_LD) {
677 		pr_cont("Load Error Report\n");
678 		recoverable = 0;
679 	}
680 
681 	if (reason & MCSR_ST) {
682 		pr_cont("Store Error Report\n");
683 		recoverable = 0;
684 	}
685 
686 	if (reason & MCSR_LDG) {
687 		pr_cont("Guarded Load Error Report\n");
688 		recoverable = 0;
689 	}
690 
691 	if (reason & MCSR_TLBSYNC)
692 		pr_cont("Simultaneous tlbsync operations\n");
693 
694 	if (reason & MCSR_BSL2_ERR) {
695 		pr_cont("Level 2 Cache Error\n");
696 		recoverable = 0;
697 	}
698 
699 	if (reason & MCSR_MAV) {
700 		u64 addr;
701 
702 		addr = mfspr(SPRN_MCAR);
703 		addr |= (u64)mfspr(SPRN_MCARU) << 32;
704 
705 		pr_cont("Machine Check %s Address: %#llx\n",
706 		       reason & MCSR_MEA ? "Effective" : "Physical", addr);
707 	}
708 
709 silent_out:
710 	mtspr(SPRN_MCSR, mcsr);
711 	return mfspr(SPRN_MCSR) == 0 && recoverable;
712 }
713 
714 int machine_check_e500(struct pt_regs *regs)
715 {
716 	unsigned long reason = mfspr(SPRN_MCSR);
717 
718 	if (reason & MCSR_BUS_RBERR) {
719 		if (fsl_rio_mcheck_exception(regs))
720 			return 1;
721 		if (fsl_pci_mcheck_exception(regs))
722 			return 1;
723 	}
724 
725 	printk("Machine check in kernel mode.\n");
726 	printk("Caused by (from MCSR=%lx): ", reason);
727 
728 	if (reason & MCSR_MCP)
729 		pr_cont("Machine Check Signal\n");
730 	if (reason & MCSR_ICPERR)
731 		pr_cont("Instruction Cache Parity Error\n");
732 	if (reason & MCSR_DCP_PERR)
733 		pr_cont("Data Cache Push Parity Error\n");
734 	if (reason & MCSR_DCPERR)
735 		pr_cont("Data Cache Parity Error\n");
736 	if (reason & MCSR_BUS_IAERR)
737 		pr_cont("Bus - Instruction Address Error\n");
738 	if (reason & MCSR_BUS_RAERR)
739 		pr_cont("Bus - Read Address Error\n");
740 	if (reason & MCSR_BUS_WAERR)
741 		pr_cont("Bus - Write Address Error\n");
742 	if (reason & MCSR_BUS_IBERR)
743 		pr_cont("Bus - Instruction Data Error\n");
744 	if (reason & MCSR_BUS_RBERR)
745 		pr_cont("Bus - Read Data Bus Error\n");
746 	if (reason & MCSR_BUS_WBERR)
747 		pr_cont("Bus - Write Data Bus Error\n");
748 	if (reason & MCSR_BUS_IPERR)
749 		pr_cont("Bus - Instruction Parity Error\n");
750 	if (reason & MCSR_BUS_RPERR)
751 		pr_cont("Bus - Read Parity Error\n");
752 
753 	return 0;
754 }
755 
756 int machine_check_generic(struct pt_regs *regs)
757 {
758 	return 0;
759 }
760 #elif defined(CONFIG_E200)
761 int machine_check_e200(struct pt_regs *regs)
762 {
763 	unsigned long reason = mfspr(SPRN_MCSR);
764 
765 	printk("Machine check in kernel mode.\n");
766 	printk("Caused by (from MCSR=%lx): ", reason);
767 
768 	if (reason & MCSR_MCP)
769 		pr_cont("Machine Check Signal\n");
770 	if (reason & MCSR_CP_PERR)
771 		pr_cont("Cache Push Parity Error\n");
772 	if (reason & MCSR_CPERR)
773 		pr_cont("Cache Parity Error\n");
774 	if (reason & MCSR_EXCP_ERR)
775 		pr_cont("ISI, ITLB, or Bus Error on first instruction fetch for an exception handler\n");
776 	if (reason & MCSR_BUS_IRERR)
777 		pr_cont("Bus - Read Bus Error on instruction fetch\n");
778 	if (reason & MCSR_BUS_DRERR)
779 		pr_cont("Bus - Read Bus Error on data load\n");
780 	if (reason & MCSR_BUS_WRERR)
781 		pr_cont("Bus - Write Bus Error on buffered store or cache line push\n");
782 
783 	return 0;
784 }
785 #elif defined(CONFIG_PPC32)
786 int machine_check_generic(struct pt_regs *regs)
787 {
788 	unsigned long reason = regs->msr;
789 
790 	printk("Machine check in kernel mode.\n");
791 	printk("Caused by (from SRR1=%lx): ", reason);
792 	switch (reason & 0x601F0000) {
793 	case 0x80000:
794 		pr_cont("Machine check signal\n");
795 		break;
796 	case 0:		/* for 601 */
797 	case 0x40000:
798 	case 0x140000:	/* 7450 MSS error and TEA */
799 		pr_cont("Transfer error ack signal\n");
800 		break;
801 	case 0x20000:
802 		pr_cont("Data parity error signal\n");
803 		break;
804 	case 0x10000:
805 		pr_cont("Address parity error signal\n");
806 		break;
807 	case 0x20000000:
808 		pr_cont("L1 Data Cache error\n");
809 		break;
810 	case 0x40000000:
811 		pr_cont("L1 Instruction Cache error\n");
812 		break;
813 	case 0x00100000:
814 		pr_cont("L2 data cache parity error\n");
815 		break;
816 	default:
817 		pr_cont("Unknown values in msr\n");
818 	}
819 	return 0;
820 }
821 #endif /* everything else */
822 
823 void machine_check_exception(struct pt_regs *regs)
824 {
825 	int recover = 0;
826 	bool nested = in_nmi();
827 	if (!nested)
828 		nmi_enter();
829 
830 	__this_cpu_inc(irq_stat.mce_exceptions);
831 
832 	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE);
833 
834 	/* See if any machine dependent calls. In theory, we would want
835 	 * to call the CPU first, and call the ppc_md. one if the CPU
836 	 * one returns a positive number. However there is existing code
837 	 * that assumes the board gets a first chance, so let's keep it
838 	 * that way for now and fix things later. --BenH.
839 	 */
840 	if (ppc_md.machine_check_exception)
841 		recover = ppc_md.machine_check_exception(regs);
842 	else if (cur_cpu_spec->machine_check)
843 		recover = cur_cpu_spec->machine_check(regs);
844 
845 	if (recover > 0)
846 		goto bail;
847 
848 	if (debugger_fault_handler(regs))
849 		goto bail;
850 
851 	if (check_io_access(regs))
852 		goto bail;
853 
854 	if (!nested)
855 		nmi_exit();
856 
857 	die("Machine check", regs, SIGBUS);
858 
859 	/* Must die if the interrupt is not recoverable */
860 	if (!(regs->msr & MSR_RI))
861 		nmi_panic(regs, "Unrecoverable Machine check");
862 
863 	return;
864 
865 bail:
866 	if (!nested)
867 		nmi_exit();
868 }
869 
870 void SMIException(struct pt_regs *regs)
871 {
872 	die("System Management Interrupt", regs, SIGABRT);
873 }
874 
875 #ifdef CONFIG_VSX
876 static void p9_hmi_special_emu(struct pt_regs *regs)
877 {
878 	unsigned int ra, rb, t, i, sel, instr, rc;
879 	const void __user *addr;
880 	u8 vbuf[16], *vdst;
881 	unsigned long ea, msr, msr_mask;
882 	bool swap;
883 
884 	if (__get_user_inatomic(instr, (unsigned int __user *)regs->nip))
885 		return;
886 
887 	/*
888 	 * lxvb16x	opcode: 0x7c0006d8
889 	 * lxvd2x	opcode: 0x7c000698
890 	 * lxvh8x	opcode: 0x7c000658
891 	 * lxvw4x	opcode: 0x7c000618
892 	 */
893 	if ((instr & 0xfc00073e) != 0x7c000618) {
894 		pr_devel("HMI vec emu: not vector CI %i:%s[%d] nip=%016lx"
895 			 " instr=%08x\n",
896 			 smp_processor_id(), current->comm, current->pid,
897 			 regs->nip, instr);
898 		return;
899 	}
900 
901 	/* Grab vector registers into the task struct */
902 	msr = regs->msr; /* Grab msr before we flush the bits */
903 	flush_vsx_to_thread(current);
904 	enable_kernel_altivec();
905 
906 	/*
907 	 * Is userspace running with a different endian (this is rare but
908 	 * not impossible)
909 	 */
910 	swap = (msr & MSR_LE) != (MSR_KERNEL & MSR_LE);
911 
912 	/* Decode the instruction */
913 	ra = (instr >> 16) & 0x1f;
914 	rb = (instr >> 11) & 0x1f;
915 	t = (instr >> 21) & 0x1f;
916 	if (instr & 1)
917 		vdst = (u8 *)&current->thread.vr_state.vr[t];
918 	else
919 		vdst = (u8 *)&current->thread.fp_state.fpr[t][0];
920 
921 	/* Grab the vector address */
922 	ea = regs->gpr[rb] + (ra ? regs->gpr[ra] : 0);
923 	if (is_32bit_task())
924 		ea &= 0xfffffffful;
925 	addr = (__force const void __user *)ea;
926 
927 	/* Check it */
928 	if (!access_ok(addr, 16)) {
929 		pr_devel("HMI vec emu: bad access %i:%s[%d] nip=%016lx"
930 			 " instr=%08x addr=%016lx\n",
931 			 smp_processor_id(), current->comm, current->pid,
932 			 regs->nip, instr, (unsigned long)addr);
933 		return;
934 	}
935 
936 	/* Read the vector */
937 	rc = 0;
938 	if ((unsigned long)addr & 0xfUL)
939 		/* unaligned case */
940 		rc = __copy_from_user_inatomic(vbuf, addr, 16);
941 	else
942 		__get_user_atomic_128_aligned(vbuf, addr, rc);
943 	if (rc) {
944 		pr_devel("HMI vec emu: page fault %i:%s[%d] nip=%016lx"
945 			 " instr=%08x addr=%016lx\n",
946 			 smp_processor_id(), current->comm, current->pid,
947 			 regs->nip, instr, (unsigned long)addr);
948 		return;
949 	}
950 
951 	pr_devel("HMI vec emu: emulated vector CI %i:%s[%d] nip=%016lx"
952 		 " instr=%08x addr=%016lx\n",
953 		 smp_processor_id(), current->comm, current->pid, regs->nip,
954 		 instr, (unsigned long) addr);
955 
956 	/* Grab instruction "selector" */
957 	sel = (instr >> 6) & 3;
958 
959 	/*
960 	 * Check to make sure the facility is actually enabled. This
961 	 * could happen if we get a false positive hit.
962 	 *
963 	 * lxvd2x/lxvw4x always check MSR VSX sel = 0,2
964 	 * lxvh8x/lxvb16x check MSR VSX or VEC depending on VSR used sel = 1,3
965 	 */
966 	msr_mask = MSR_VSX;
967 	if ((sel & 1) && (instr & 1)) /* lxvh8x & lxvb16x + VSR >= 32 */
968 		msr_mask = MSR_VEC;
969 	if (!(msr & msr_mask)) {
970 		pr_devel("HMI vec emu: MSR fac clear %i:%s[%d] nip=%016lx"
971 			 " instr=%08x msr:%016lx\n",
972 			 smp_processor_id(), current->comm, current->pid,
973 			 regs->nip, instr, msr);
974 		return;
975 	}
976 
977 	/* Do logging here before we modify sel based on endian */
978 	switch (sel) {
979 	case 0:	/* lxvw4x */
980 		PPC_WARN_EMULATED(lxvw4x, regs);
981 		break;
982 	case 1: /* lxvh8x */
983 		PPC_WARN_EMULATED(lxvh8x, regs);
984 		break;
985 	case 2: /* lxvd2x */
986 		PPC_WARN_EMULATED(lxvd2x, regs);
987 		break;
988 	case 3: /* lxvb16x */
989 		PPC_WARN_EMULATED(lxvb16x, regs);
990 		break;
991 	}
992 
993 #ifdef __LITTLE_ENDIAN__
994 	/*
995 	 * An LE kernel stores the vector in the task struct as an LE
996 	 * byte array (effectively swapping both the components and
997 	 * the content of the components). Those instructions expect
998 	 * the components to remain in ascending address order, so we
999 	 * swap them back.
1000 	 *
1001 	 * If we are running a BE user space, the expectation is that
1002 	 * of a simple memcpy, so forcing the emulation to look like
1003 	 * a lxvb16x should do the trick.
1004 	 */
1005 	if (swap)
1006 		sel = 3;
1007 
1008 	switch (sel) {
1009 	case 0:	/* lxvw4x */
1010 		for (i = 0; i < 4; i++)
1011 			((u32 *)vdst)[i] = ((u32 *)vbuf)[3-i];
1012 		break;
1013 	case 1: /* lxvh8x */
1014 		for (i = 0; i < 8; i++)
1015 			((u16 *)vdst)[i] = ((u16 *)vbuf)[7-i];
1016 		break;
1017 	case 2: /* lxvd2x */
1018 		for (i = 0; i < 2; i++)
1019 			((u64 *)vdst)[i] = ((u64 *)vbuf)[1-i];
1020 		break;
1021 	case 3: /* lxvb16x */
1022 		for (i = 0; i < 16; i++)
1023 			vdst[i] = vbuf[15-i];
1024 		break;
1025 	}
1026 #else /* __LITTLE_ENDIAN__ */
1027 	/* On a big endian kernel, a BE userspace only needs a memcpy */
1028 	if (!swap)
1029 		sel = 3;
1030 
1031 	/* Otherwise, we need to swap the content of the components */
1032 	switch (sel) {
1033 	case 0:	/* lxvw4x */
1034 		for (i = 0; i < 4; i++)
1035 			((u32 *)vdst)[i] = cpu_to_le32(((u32 *)vbuf)[i]);
1036 		break;
1037 	case 1: /* lxvh8x */
1038 		for (i = 0; i < 8; i++)
1039 			((u16 *)vdst)[i] = cpu_to_le16(((u16 *)vbuf)[i]);
1040 		break;
1041 	case 2: /* lxvd2x */
1042 		for (i = 0; i < 2; i++)
1043 			((u64 *)vdst)[i] = cpu_to_le64(((u64 *)vbuf)[i]);
1044 		break;
1045 	case 3: /* lxvb16x */
1046 		memcpy(vdst, vbuf, 16);
1047 		break;
1048 	}
1049 #endif /* !__LITTLE_ENDIAN__ */
1050 
1051 	/* Go to next instruction */
1052 	regs->nip += 4;
1053 }
1054 #endif /* CONFIG_VSX */
1055 
1056 void handle_hmi_exception(struct pt_regs *regs)
1057 {
1058 	struct pt_regs *old_regs;
1059 
1060 	old_regs = set_irq_regs(regs);
1061 	irq_enter();
1062 
1063 #ifdef CONFIG_VSX
1064 	/* Real mode flagged P9 special emu is needed */
1065 	if (local_paca->hmi_p9_special_emu) {
1066 		local_paca->hmi_p9_special_emu = 0;
1067 
1068 		/*
1069 		 * We don't want to take page faults while doing the
1070 		 * emulation, we just replay the instruction if necessary.
1071 		 */
1072 		pagefault_disable();
1073 		p9_hmi_special_emu(regs);
1074 		pagefault_enable();
1075 	}
1076 #endif /* CONFIG_VSX */
1077 
1078 	if (ppc_md.handle_hmi_exception)
1079 		ppc_md.handle_hmi_exception(regs);
1080 
1081 	irq_exit();
1082 	set_irq_regs(old_regs);
1083 }
1084 
1085 void unknown_exception(struct pt_regs *regs)
1086 {
1087 	enum ctx_state prev_state = exception_enter();
1088 
1089 	printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
1090 	       regs->nip, regs->msr, regs->trap);
1091 
1092 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1093 
1094 	exception_exit(prev_state);
1095 }
1096 
1097 void instruction_breakpoint_exception(struct pt_regs *regs)
1098 {
1099 	enum ctx_state prev_state = exception_enter();
1100 
1101 	if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
1102 					5, SIGTRAP) == NOTIFY_STOP)
1103 		goto bail;
1104 	if (debugger_iabr_match(regs))
1105 		goto bail;
1106 	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1107 
1108 bail:
1109 	exception_exit(prev_state);
1110 }
1111 
1112 void RunModeException(struct pt_regs *regs)
1113 {
1114 	_exception(SIGTRAP, regs, TRAP_UNK, 0);
1115 }
1116 
1117 void single_step_exception(struct pt_regs *regs)
1118 {
1119 	enum ctx_state prev_state = exception_enter();
1120 
1121 	clear_single_step(regs);
1122 	clear_br_trace(regs);
1123 
1124 	if (kprobe_post_handler(regs))
1125 		return;
1126 
1127 	if (notify_die(DIE_SSTEP, "single_step", regs, 5,
1128 					5, SIGTRAP) == NOTIFY_STOP)
1129 		goto bail;
1130 	if (debugger_sstep(regs))
1131 		goto bail;
1132 
1133 	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
1134 
1135 bail:
1136 	exception_exit(prev_state);
1137 }
1138 NOKPROBE_SYMBOL(single_step_exception);
1139 
1140 /*
1141  * After we have successfully emulated an instruction, we have to
1142  * check if the instruction was being single-stepped, and if so,
1143  * pretend we got a single-step exception.  This was pointed out
1144  * by Kumar Gala.  -- paulus
1145  */
1146 static void emulate_single_step(struct pt_regs *regs)
1147 {
1148 	if (single_stepping(regs))
1149 		single_step_exception(regs);
1150 }
1151 
1152 static inline int __parse_fpscr(unsigned long fpscr)
1153 {
1154 	int ret = FPE_FLTUNK;
1155 
1156 	/* Invalid operation */
1157 	if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))
1158 		ret = FPE_FLTINV;
1159 
1160 	/* Overflow */
1161 	else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX))
1162 		ret = FPE_FLTOVF;
1163 
1164 	/* Underflow */
1165 	else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX))
1166 		ret = FPE_FLTUND;
1167 
1168 	/* Divide by zero */
1169 	else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX))
1170 		ret = FPE_FLTDIV;
1171 
1172 	/* Inexact result */
1173 	else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX))
1174 		ret = FPE_FLTRES;
1175 
1176 	return ret;
1177 }
1178 
1179 static void parse_fpe(struct pt_regs *regs)
1180 {
1181 	int code = 0;
1182 
1183 	flush_fp_to_thread(current);
1184 
1185 	code = __parse_fpscr(current->thread.fp_state.fpscr);
1186 
1187 	_exception(SIGFPE, regs, code, regs->nip);
1188 }
1189 
1190 /*
1191  * Illegal instruction emulation support.  Originally written to
1192  * provide the PVR to user applications using the mfspr rd, PVR.
1193  * Return non-zero if we can't emulate, or -EFAULT if the associated
1194  * memory access caused an access fault.  Return zero on success.
1195  *
1196  * There are a couple of ways to do this, either "decode" the instruction
1197  * or directly match lots of bits.  In this case, matching lots of
1198  * bits is faster and easier.
1199  *
1200  */
1201 static int emulate_string_inst(struct pt_regs *regs, u32 instword)
1202 {
1203 	u8 rT = (instword >> 21) & 0x1f;
1204 	u8 rA = (instword >> 16) & 0x1f;
1205 	u8 NB_RB = (instword >> 11) & 0x1f;
1206 	u32 num_bytes;
1207 	unsigned long EA;
1208 	int pos = 0;
1209 
1210 	/* Early out if we are an invalid form of lswx */
1211 	if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX)
1212 		if ((rT == rA) || (rT == NB_RB))
1213 			return -EINVAL;
1214 
1215 	EA = (rA == 0) ? 0 : regs->gpr[rA];
1216 
1217 	switch (instword & PPC_INST_STRING_MASK) {
1218 		case PPC_INST_LSWX:
1219 		case PPC_INST_STSWX:
1220 			EA += NB_RB;
1221 			num_bytes = regs->xer & 0x7f;
1222 			break;
1223 		case PPC_INST_LSWI:
1224 		case PPC_INST_STSWI:
1225 			num_bytes = (NB_RB == 0) ? 32 : NB_RB;
1226 			break;
1227 		default:
1228 			return -EINVAL;
1229 	}
1230 
1231 	while (num_bytes != 0)
1232 	{
1233 		u8 val;
1234 		u32 shift = 8 * (3 - (pos & 0x3));
1235 
1236 		/* if process is 32-bit, clear upper 32 bits of EA */
1237 		if ((regs->msr & MSR_64BIT) == 0)
1238 			EA &= 0xFFFFFFFF;
1239 
1240 		switch ((instword & PPC_INST_STRING_MASK)) {
1241 			case PPC_INST_LSWX:
1242 			case PPC_INST_LSWI:
1243 				if (get_user(val, (u8 __user *)EA))
1244 					return -EFAULT;
1245 				/* first time updating this reg,
1246 				 * zero it out */
1247 				if (pos == 0)
1248 					regs->gpr[rT] = 0;
1249 				regs->gpr[rT] |= val << shift;
1250 				break;
1251 			case PPC_INST_STSWI:
1252 			case PPC_INST_STSWX:
1253 				val = regs->gpr[rT] >> shift;
1254 				if (put_user(val, (u8 __user *)EA))
1255 					return -EFAULT;
1256 				break;
1257 		}
1258 		/* move EA to next address */
1259 		EA += 1;
1260 		num_bytes--;
1261 
1262 		/* manage our position within the register */
1263 		if (++pos == 4) {
1264 			pos = 0;
1265 			if (++rT == 32)
1266 				rT = 0;
1267 		}
1268 	}
1269 
1270 	return 0;
1271 }
1272 
1273 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword)
1274 {
1275 	u32 ra,rs;
1276 	unsigned long tmp;
1277 
1278 	ra = (instword >> 16) & 0x1f;
1279 	rs = (instword >> 21) & 0x1f;
1280 
1281 	tmp = regs->gpr[rs];
1282 	tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL);
1283 	tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL);
1284 	tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL;
1285 	regs->gpr[ra] = tmp;
1286 
1287 	return 0;
1288 }
1289 
1290 static int emulate_isel(struct pt_regs *regs, u32 instword)
1291 {
1292 	u8 rT = (instword >> 21) & 0x1f;
1293 	u8 rA = (instword >> 16) & 0x1f;
1294 	u8 rB = (instword >> 11) & 0x1f;
1295 	u8 BC = (instword >> 6) & 0x1f;
1296 	u8 bit;
1297 	unsigned long tmp;
1298 
1299 	tmp = (rA == 0) ? 0 : regs->gpr[rA];
1300 	bit = (regs->ccr >> (31 - BC)) & 0x1;
1301 
1302 	regs->gpr[rT] = bit ? tmp : regs->gpr[rB];
1303 
1304 	return 0;
1305 }
1306 
1307 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1308 static inline bool tm_abort_check(struct pt_regs *regs, int cause)
1309 {
1310         /* If we're emulating a load/store in an active transaction, we cannot
1311          * emulate it as the kernel operates in transaction suspended context.
1312          * We need to abort the transaction.  This creates a persistent TM
1313          * abort so tell the user what caused it with a new code.
1314 	 */
1315 	if (MSR_TM_TRANSACTIONAL(regs->msr)) {
1316 		tm_enable();
1317 		tm_abort(cause);
1318 		return true;
1319 	}
1320 	return false;
1321 }
1322 #else
1323 static inline bool tm_abort_check(struct pt_regs *regs, int reason)
1324 {
1325 	return false;
1326 }
1327 #endif
1328 
1329 static int emulate_instruction(struct pt_regs *regs)
1330 {
1331 	u32 instword;
1332 	u32 rd;
1333 
1334 	if (!user_mode(regs))
1335 		return -EINVAL;
1336 	CHECK_FULL_REGS(regs);
1337 
1338 	if (get_user(instword, (u32 __user *)(regs->nip)))
1339 		return -EFAULT;
1340 
1341 	/* Emulate the mfspr rD, PVR. */
1342 	if ((instword & PPC_INST_MFSPR_PVR_MASK) == PPC_INST_MFSPR_PVR) {
1343 		PPC_WARN_EMULATED(mfpvr, regs);
1344 		rd = (instword >> 21) & 0x1f;
1345 		regs->gpr[rd] = mfspr(SPRN_PVR);
1346 		return 0;
1347 	}
1348 
1349 	/* Emulating the dcba insn is just a no-op.  */
1350 	if ((instword & PPC_INST_DCBA_MASK) == PPC_INST_DCBA) {
1351 		PPC_WARN_EMULATED(dcba, regs);
1352 		return 0;
1353 	}
1354 
1355 	/* Emulate the mcrxr insn.  */
1356 	if ((instword & PPC_INST_MCRXR_MASK) == PPC_INST_MCRXR) {
1357 		int shift = (instword >> 21) & 0x1c;
1358 		unsigned long msk = 0xf0000000UL >> shift;
1359 
1360 		PPC_WARN_EMULATED(mcrxr, regs);
1361 		regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk);
1362 		regs->xer &= ~0xf0000000UL;
1363 		return 0;
1364 	}
1365 
1366 	/* Emulate load/store string insn. */
1367 	if ((instword & PPC_INST_STRING_GEN_MASK) == PPC_INST_STRING) {
1368 		if (tm_abort_check(regs,
1369 				   TM_CAUSE_EMULATE | TM_CAUSE_PERSISTENT))
1370 			return -EINVAL;
1371 		PPC_WARN_EMULATED(string, regs);
1372 		return emulate_string_inst(regs, instword);
1373 	}
1374 
1375 	/* Emulate the popcntb (Population Count Bytes) instruction. */
1376 	if ((instword & PPC_INST_POPCNTB_MASK) == PPC_INST_POPCNTB) {
1377 		PPC_WARN_EMULATED(popcntb, regs);
1378 		return emulate_popcntb_inst(regs, instword);
1379 	}
1380 
1381 	/* Emulate isel (Integer Select) instruction */
1382 	if ((instword & PPC_INST_ISEL_MASK) == PPC_INST_ISEL) {
1383 		PPC_WARN_EMULATED(isel, regs);
1384 		return emulate_isel(regs, instword);
1385 	}
1386 
1387 	/* Emulate sync instruction variants */
1388 	if ((instword & PPC_INST_SYNC_MASK) == PPC_INST_SYNC) {
1389 		PPC_WARN_EMULATED(sync, regs);
1390 		asm volatile("sync");
1391 		return 0;
1392 	}
1393 
1394 #ifdef CONFIG_PPC64
1395 	/* Emulate the mfspr rD, DSCR. */
1396 	if ((((instword & PPC_INST_MFSPR_DSCR_USER_MASK) ==
1397 		PPC_INST_MFSPR_DSCR_USER) ||
1398 	     ((instword & PPC_INST_MFSPR_DSCR_MASK) ==
1399 		PPC_INST_MFSPR_DSCR)) &&
1400 			cpu_has_feature(CPU_FTR_DSCR)) {
1401 		PPC_WARN_EMULATED(mfdscr, regs);
1402 		rd = (instword >> 21) & 0x1f;
1403 		regs->gpr[rd] = mfspr(SPRN_DSCR);
1404 		return 0;
1405 	}
1406 	/* Emulate the mtspr DSCR, rD. */
1407 	if ((((instword & PPC_INST_MTSPR_DSCR_USER_MASK) ==
1408 		PPC_INST_MTSPR_DSCR_USER) ||
1409 	     ((instword & PPC_INST_MTSPR_DSCR_MASK) ==
1410 		PPC_INST_MTSPR_DSCR)) &&
1411 			cpu_has_feature(CPU_FTR_DSCR)) {
1412 		PPC_WARN_EMULATED(mtdscr, regs);
1413 		rd = (instword >> 21) & 0x1f;
1414 		current->thread.dscr = regs->gpr[rd];
1415 		current->thread.dscr_inherit = 1;
1416 		mtspr(SPRN_DSCR, current->thread.dscr);
1417 		return 0;
1418 	}
1419 #endif
1420 
1421 	return -EINVAL;
1422 }
1423 
1424 int is_valid_bugaddr(unsigned long addr)
1425 {
1426 	return is_kernel_addr(addr);
1427 }
1428 
1429 #ifdef CONFIG_MATH_EMULATION
1430 static int emulate_math(struct pt_regs *regs)
1431 {
1432 	int ret;
1433 	extern int do_mathemu(struct pt_regs *regs);
1434 
1435 	ret = do_mathemu(regs);
1436 	if (ret >= 0)
1437 		PPC_WARN_EMULATED(math, regs);
1438 
1439 	switch (ret) {
1440 	case 0:
1441 		emulate_single_step(regs);
1442 		return 0;
1443 	case 1: {
1444 			int code = 0;
1445 			code = __parse_fpscr(current->thread.fp_state.fpscr);
1446 			_exception(SIGFPE, regs, code, regs->nip);
1447 			return 0;
1448 		}
1449 	case -EFAULT:
1450 		_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1451 		return 0;
1452 	}
1453 
1454 	return -1;
1455 }
1456 #else
1457 static inline int emulate_math(struct pt_regs *regs) { return -1; }
1458 #endif
1459 
1460 void program_check_exception(struct pt_regs *regs)
1461 {
1462 	enum ctx_state prev_state = exception_enter();
1463 	unsigned int reason = get_reason(regs);
1464 
1465 	/* We can now get here via a FP Unavailable exception if the core
1466 	 * has no FPU, in that case the reason flags will be 0 */
1467 
1468 	if (reason & REASON_FP) {
1469 		/* IEEE FP exception */
1470 		parse_fpe(regs);
1471 		goto bail;
1472 	}
1473 	if (reason & REASON_TRAP) {
1474 		unsigned long bugaddr;
1475 		/* Debugger is first in line to stop recursive faults in
1476 		 * rcu_lock, notify_die, or atomic_notifier_call_chain */
1477 		if (debugger_bpt(regs))
1478 			goto bail;
1479 
1480 		if (kprobe_handler(regs))
1481 			goto bail;
1482 
1483 		/* trap exception */
1484 		if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP)
1485 				== NOTIFY_STOP)
1486 			goto bail;
1487 
1488 		bugaddr = regs->nip;
1489 		/*
1490 		 * Fixup bugaddr for BUG_ON() in real mode
1491 		 */
1492 		if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR))
1493 			bugaddr += PAGE_OFFSET;
1494 
1495 		if (!(regs->msr & MSR_PR) &&  /* not user-mode */
1496 		    report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) {
1497 			regs->nip += 4;
1498 			goto bail;
1499 		}
1500 		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
1501 		goto bail;
1502 	}
1503 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1504 	if (reason & REASON_TM) {
1505 		/* This is a TM "Bad Thing Exception" program check.
1506 		 * This occurs when:
1507 		 * -  An rfid/hrfid/mtmsrd attempts to cause an illegal
1508 		 *    transition in TM states.
1509 		 * -  A trechkpt is attempted when transactional.
1510 		 * -  A treclaim is attempted when non transactional.
1511 		 * -  A tend is illegally attempted.
1512 		 * -  writing a TM SPR when transactional.
1513 		 *
1514 		 * If usermode caused this, it's done something illegal and
1515 		 * gets a SIGILL slap on the wrist.  We call it an illegal
1516 		 * operand to distinguish from the instruction just being bad
1517 		 * (e.g. executing a 'tend' on a CPU without TM!); it's an
1518 		 * illegal /placement/ of a valid instruction.
1519 		 */
1520 		if (user_mode(regs)) {
1521 			_exception(SIGILL, regs, ILL_ILLOPN, regs->nip);
1522 			goto bail;
1523 		} else {
1524 			printk(KERN_EMERG "Unexpected TM Bad Thing exception "
1525 			       "at %lx (msr 0x%lx) tm_scratch=%llx\n",
1526 			       regs->nip, regs->msr, get_paca()->tm_scratch);
1527 			die("Unrecoverable exception", regs, SIGABRT);
1528 		}
1529 	}
1530 #endif
1531 
1532 	/*
1533 	 * If we took the program check in the kernel skip down to sending a
1534 	 * SIGILL. The subsequent cases all relate to emulating instructions
1535 	 * which we should only do for userspace. We also do not want to enable
1536 	 * interrupts for kernel faults because that might lead to further
1537 	 * faults, and loose the context of the original exception.
1538 	 */
1539 	if (!user_mode(regs))
1540 		goto sigill;
1541 
1542 	/* We restore the interrupt state now */
1543 	if (!arch_irq_disabled_regs(regs))
1544 		local_irq_enable();
1545 
1546 	/* (reason & REASON_ILLEGAL) would be the obvious thing here,
1547 	 * but there seems to be a hardware bug on the 405GP (RevD)
1548 	 * that means ESR is sometimes set incorrectly - either to
1549 	 * ESR_DST (!?) or 0.  In the process of chasing this with the
1550 	 * hardware people - not sure if it can happen on any illegal
1551 	 * instruction or only on FP instructions, whether there is a
1552 	 * pattern to occurrences etc. -dgibson 31/Mar/2003
1553 	 */
1554 	if (!emulate_math(regs))
1555 		goto bail;
1556 
1557 	/* Try to emulate it if we should. */
1558 	if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
1559 		switch (emulate_instruction(regs)) {
1560 		case 0:
1561 			regs->nip += 4;
1562 			emulate_single_step(regs);
1563 			goto bail;
1564 		case -EFAULT:
1565 			_exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
1566 			goto bail;
1567 		}
1568 	}
1569 
1570 sigill:
1571 	if (reason & REASON_PRIVILEGED)
1572 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
1573 	else
1574 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1575 
1576 bail:
1577 	exception_exit(prev_state);
1578 }
1579 NOKPROBE_SYMBOL(program_check_exception);
1580 
1581 /*
1582  * This occurs when running in hypervisor mode on POWER6 or later
1583  * and an illegal instruction is encountered.
1584  */
1585 void emulation_assist_interrupt(struct pt_regs *regs)
1586 {
1587 	regs->msr |= REASON_ILLEGAL;
1588 	program_check_exception(regs);
1589 }
1590 NOKPROBE_SYMBOL(emulation_assist_interrupt);
1591 
1592 void alignment_exception(struct pt_regs *regs)
1593 {
1594 	enum ctx_state prev_state = exception_enter();
1595 	int sig, code, fixed = 0;
1596 
1597 	/* We restore the interrupt state now */
1598 	if (!arch_irq_disabled_regs(regs))
1599 		local_irq_enable();
1600 
1601 	if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT))
1602 		goto bail;
1603 
1604 	/* we don't implement logging of alignment exceptions */
1605 	if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS))
1606 		fixed = fix_alignment(regs);
1607 
1608 	if (fixed == 1) {
1609 		regs->nip += 4;	/* skip over emulated instruction */
1610 		emulate_single_step(regs);
1611 		goto bail;
1612 	}
1613 
1614 	/* Operand address was bad */
1615 	if (fixed == -EFAULT) {
1616 		sig = SIGSEGV;
1617 		code = SEGV_ACCERR;
1618 	} else {
1619 		sig = SIGBUS;
1620 		code = BUS_ADRALN;
1621 	}
1622 	if (user_mode(regs))
1623 		_exception(sig, regs, code, regs->dar);
1624 	else
1625 		bad_page_fault(regs, regs->dar, sig);
1626 
1627 bail:
1628 	exception_exit(prev_state);
1629 }
1630 
1631 void StackOverflow(struct pt_regs *regs)
1632 {
1633 	pr_crit("Kernel stack overflow in process %s[%d], r1=%lx\n",
1634 		current->comm, task_pid_nr(current), regs->gpr[1]);
1635 	debugger(regs);
1636 	show_regs(regs);
1637 	panic("kernel stack overflow");
1638 }
1639 
1640 void kernel_fp_unavailable_exception(struct pt_regs *regs)
1641 {
1642 	enum ctx_state prev_state = exception_enter();
1643 
1644 	printk(KERN_EMERG "Unrecoverable FP Unavailable Exception "
1645 			  "%lx at %lx\n", regs->trap, regs->nip);
1646 	die("Unrecoverable FP Unavailable Exception", regs, SIGABRT);
1647 
1648 	exception_exit(prev_state);
1649 }
1650 
1651 void altivec_unavailable_exception(struct pt_regs *regs)
1652 {
1653 	enum ctx_state prev_state = exception_enter();
1654 
1655 	if (user_mode(regs)) {
1656 		/* A user program has executed an altivec instruction,
1657 		   but this kernel doesn't support altivec. */
1658 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1659 		goto bail;
1660 	}
1661 
1662 	printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception "
1663 			"%lx at %lx\n", regs->trap, regs->nip);
1664 	die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT);
1665 
1666 bail:
1667 	exception_exit(prev_state);
1668 }
1669 
1670 void vsx_unavailable_exception(struct pt_regs *regs)
1671 {
1672 	if (user_mode(regs)) {
1673 		/* A user program has executed an vsx instruction,
1674 		   but this kernel doesn't support vsx. */
1675 		_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1676 		return;
1677 	}
1678 
1679 	printk(KERN_EMERG "Unrecoverable VSX Unavailable Exception "
1680 			"%lx at %lx\n", regs->trap, regs->nip);
1681 	die("Unrecoverable VSX Unavailable Exception", regs, SIGABRT);
1682 }
1683 
1684 #ifdef CONFIG_PPC64
1685 static void tm_unavailable(struct pt_regs *regs)
1686 {
1687 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1688 	if (user_mode(regs)) {
1689 		current->thread.load_tm++;
1690 		regs->msr |= MSR_TM;
1691 		tm_enable();
1692 		tm_restore_sprs(&current->thread);
1693 		return;
1694 	}
1695 #endif
1696 	pr_emerg("Unrecoverable TM Unavailable Exception "
1697 			"%lx at %lx\n", regs->trap, regs->nip);
1698 	die("Unrecoverable TM Unavailable Exception", regs, SIGABRT);
1699 }
1700 
1701 void facility_unavailable_exception(struct pt_regs *regs)
1702 {
1703 	static char *facility_strings[] = {
1704 		[FSCR_FP_LG] = "FPU",
1705 		[FSCR_VECVSX_LG] = "VMX/VSX",
1706 		[FSCR_DSCR_LG] = "DSCR",
1707 		[FSCR_PM_LG] = "PMU SPRs",
1708 		[FSCR_BHRB_LG] = "BHRB",
1709 		[FSCR_TM_LG] = "TM",
1710 		[FSCR_EBB_LG] = "EBB",
1711 		[FSCR_TAR_LG] = "TAR",
1712 		[FSCR_MSGP_LG] = "MSGP",
1713 		[FSCR_SCV_LG] = "SCV",
1714 	};
1715 	char *facility = "unknown";
1716 	u64 value;
1717 	u32 instword, rd;
1718 	u8 status;
1719 	bool hv;
1720 
1721 	hv = (TRAP(regs) == 0xf80);
1722 	if (hv)
1723 		value = mfspr(SPRN_HFSCR);
1724 	else
1725 		value = mfspr(SPRN_FSCR);
1726 
1727 	status = value >> 56;
1728 	if ((hv || status >= 2) &&
1729 	    (status < ARRAY_SIZE(facility_strings)) &&
1730 	    facility_strings[status])
1731 		facility = facility_strings[status];
1732 
1733 	/* We should not have taken this interrupt in kernel */
1734 	if (!user_mode(regs)) {
1735 		pr_emerg("Facility '%s' unavailable (%d) exception in kernel mode at %lx\n",
1736 			 facility, status, regs->nip);
1737 		die("Unexpected facility unavailable exception", regs, SIGABRT);
1738 	}
1739 
1740 	/* We restore the interrupt state now */
1741 	if (!arch_irq_disabled_regs(regs))
1742 		local_irq_enable();
1743 
1744 	if (status == FSCR_DSCR_LG) {
1745 		/*
1746 		 * User is accessing the DSCR register using the problem
1747 		 * state only SPR number (0x03) either through a mfspr or
1748 		 * a mtspr instruction. If it is a write attempt through
1749 		 * a mtspr, then we set the inherit bit. This also allows
1750 		 * the user to write or read the register directly in the
1751 		 * future by setting via the FSCR DSCR bit. But in case it
1752 		 * is a read DSCR attempt through a mfspr instruction, we
1753 		 * just emulate the instruction instead. This code path will
1754 		 * always emulate all the mfspr instructions till the user
1755 		 * has attempted at least one mtspr instruction. This way it
1756 		 * preserves the same behaviour when the user is accessing
1757 		 * the DSCR through privilege level only SPR number (0x11)
1758 		 * which is emulated through illegal instruction exception.
1759 		 * We always leave HFSCR DSCR set.
1760 		 */
1761 		if (get_user(instword, (u32 __user *)(regs->nip))) {
1762 			pr_err("Failed to fetch the user instruction\n");
1763 			return;
1764 		}
1765 
1766 		/* Write into DSCR (mtspr 0x03, RS) */
1767 		if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK)
1768 				== PPC_INST_MTSPR_DSCR_USER) {
1769 			rd = (instword >> 21) & 0x1f;
1770 			current->thread.dscr = regs->gpr[rd];
1771 			current->thread.dscr_inherit = 1;
1772 			current->thread.fscr |= FSCR_DSCR;
1773 			mtspr(SPRN_FSCR, current->thread.fscr);
1774 		}
1775 
1776 		/* Read from DSCR (mfspr RT, 0x03) */
1777 		if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK)
1778 				== PPC_INST_MFSPR_DSCR_USER) {
1779 			if (emulate_instruction(regs)) {
1780 				pr_err("DSCR based mfspr emulation failed\n");
1781 				return;
1782 			}
1783 			regs->nip += 4;
1784 			emulate_single_step(regs);
1785 		}
1786 		return;
1787 	}
1788 
1789 	if (status == FSCR_TM_LG) {
1790 		/*
1791 		 * If we're here then the hardware is TM aware because it
1792 		 * generated an exception with FSRM_TM set.
1793 		 *
1794 		 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware
1795 		 * told us not to do TM, or the kernel is not built with TM
1796 		 * support.
1797 		 *
1798 		 * If both of those things are true, then userspace can spam the
1799 		 * console by triggering the printk() below just by continually
1800 		 * doing tbegin (or any TM instruction). So in that case just
1801 		 * send the process a SIGILL immediately.
1802 		 */
1803 		if (!cpu_has_feature(CPU_FTR_TM))
1804 			goto out;
1805 
1806 		tm_unavailable(regs);
1807 		return;
1808 	}
1809 
1810 	pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n",
1811 		hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr);
1812 
1813 out:
1814 	_exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
1815 }
1816 #endif
1817 
1818 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1819 
1820 void fp_unavailable_tm(struct pt_regs *regs)
1821 {
1822 	/* Note:  This does not handle any kind of FP laziness. */
1823 
1824 	TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n",
1825 		 regs->nip, regs->msr);
1826 
1827         /* We can only have got here if the task started using FP after
1828          * beginning the transaction.  So, the transactional regs are just a
1829          * copy of the checkpointed ones.  But, we still need to recheckpoint
1830          * as we're enabling FP for the process; it will return, abort the
1831          * transaction, and probably retry but now with FP enabled.  So the
1832          * checkpointed FP registers need to be loaded.
1833 	 */
1834 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1835 
1836 	/*
1837 	 * Reclaim initially saved out bogus (lazy) FPRs to ckfp_state, and
1838 	 * then it was overwrite by the thr->fp_state by tm_reclaim_thread().
1839 	 *
1840 	 * At this point, ck{fp,vr}_state contains the exact values we want to
1841 	 * recheckpoint.
1842 	 */
1843 
1844 	/* Enable FP for the task: */
1845 	current->thread.load_fp = 1;
1846 
1847 	/*
1848 	 * Recheckpoint all the checkpointed ckpt, ck{fp, vr}_state registers.
1849 	 */
1850 	tm_recheckpoint(&current->thread);
1851 }
1852 
1853 void altivec_unavailable_tm(struct pt_regs *regs)
1854 {
1855 	/* See the comments in fp_unavailable_tm().  This function operates
1856 	 * the same way.
1857 	 */
1858 
1859 	TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx,"
1860 		 "MSR=%lx\n",
1861 		 regs->nip, regs->msr);
1862 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1863 	current->thread.load_vec = 1;
1864 	tm_recheckpoint(&current->thread);
1865 	current->thread.used_vr = 1;
1866 }
1867 
1868 void vsx_unavailable_tm(struct pt_regs *regs)
1869 {
1870 	/* See the comments in fp_unavailable_tm().  This works similarly,
1871 	 * though we're loading both FP and VEC registers in here.
1872 	 *
1873 	 * If FP isn't in use, load FP regs.  If VEC isn't in use, load VEC
1874 	 * regs.  Either way, set MSR_VSX.
1875 	 */
1876 
1877 	TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx,"
1878 		 "MSR=%lx\n",
1879 		 regs->nip, regs->msr);
1880 
1881 	current->thread.used_vsr = 1;
1882 
1883 	/* This reclaims FP and/or VR regs if they're already enabled */
1884 	tm_reclaim_current(TM_CAUSE_FAC_UNAV);
1885 
1886 	current->thread.load_vec = 1;
1887 	current->thread.load_fp = 1;
1888 
1889 	tm_recheckpoint(&current->thread);
1890 }
1891 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1892 
1893 void performance_monitor_exception(struct pt_regs *regs)
1894 {
1895 	__this_cpu_inc(irq_stat.pmu_irqs);
1896 
1897 	perf_irq(regs);
1898 }
1899 
1900 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1901 static void handle_debug(struct pt_regs *regs, unsigned long debug_status)
1902 {
1903 	int changed = 0;
1904 	/*
1905 	 * Determine the cause of the debug event, clear the
1906 	 * event flags and send a trap to the handler. Torez
1907 	 */
1908 	if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) {
1909 		dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W);
1910 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE
1911 		current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE;
1912 #endif
1913 		do_send_trap(regs, mfspr(SPRN_DAC1), debug_status,
1914 			     5);
1915 		changed |= 0x01;
1916 	}  else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) {
1917 		dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W);
1918 		do_send_trap(regs, mfspr(SPRN_DAC2), debug_status,
1919 			     6);
1920 		changed |= 0x01;
1921 	}  else if (debug_status & DBSR_IAC1) {
1922 		current->thread.debug.dbcr0 &= ~DBCR0_IAC1;
1923 		dbcr_iac_range(current) &= ~DBCR_IAC12MODE;
1924 		do_send_trap(regs, mfspr(SPRN_IAC1), debug_status,
1925 			     1);
1926 		changed |= 0x01;
1927 	}  else if (debug_status & DBSR_IAC2) {
1928 		current->thread.debug.dbcr0 &= ~DBCR0_IAC2;
1929 		do_send_trap(regs, mfspr(SPRN_IAC2), debug_status,
1930 			     2);
1931 		changed |= 0x01;
1932 	}  else if (debug_status & DBSR_IAC3) {
1933 		current->thread.debug.dbcr0 &= ~DBCR0_IAC3;
1934 		dbcr_iac_range(current) &= ~DBCR_IAC34MODE;
1935 		do_send_trap(regs, mfspr(SPRN_IAC3), debug_status,
1936 			     3);
1937 		changed |= 0x01;
1938 	}  else if (debug_status & DBSR_IAC4) {
1939 		current->thread.debug.dbcr0 &= ~DBCR0_IAC4;
1940 		do_send_trap(regs, mfspr(SPRN_IAC4), debug_status,
1941 			     4);
1942 		changed |= 0x01;
1943 	}
1944 	/*
1945 	 * At the point this routine was called, the MSR(DE) was turned off.
1946 	 * Check all other debug flags and see if that bit needs to be turned
1947 	 * back on or not.
1948 	 */
1949 	if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
1950 			       current->thread.debug.dbcr1))
1951 		regs->msr |= MSR_DE;
1952 	else
1953 		/* Make sure the IDM flag is off */
1954 		current->thread.debug.dbcr0 &= ~DBCR0_IDM;
1955 
1956 	if (changed & 0x01)
1957 		mtspr(SPRN_DBCR0, current->thread.debug.dbcr0);
1958 }
1959 
1960 void DebugException(struct pt_regs *regs, unsigned long debug_status)
1961 {
1962 	current->thread.debug.dbsr = debug_status;
1963 
1964 	/* Hack alert: On BookE, Branch Taken stops on the branch itself, while
1965 	 * on server, it stops on the target of the branch. In order to simulate
1966 	 * the server behaviour, we thus restart right away with a single step
1967 	 * instead of stopping here when hitting a BT
1968 	 */
1969 	if (debug_status & DBSR_BT) {
1970 		regs->msr &= ~MSR_DE;
1971 
1972 		/* Disable BT */
1973 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT);
1974 		/* Clear the BT event */
1975 		mtspr(SPRN_DBSR, DBSR_BT);
1976 
1977 		/* Do the single step trick only when coming from userspace */
1978 		if (user_mode(regs)) {
1979 			current->thread.debug.dbcr0 &= ~DBCR0_BT;
1980 			current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC;
1981 			regs->msr |= MSR_DE;
1982 			return;
1983 		}
1984 
1985 		if (kprobe_post_handler(regs))
1986 			return;
1987 
1988 		if (notify_die(DIE_SSTEP, "block_step", regs, 5,
1989 			       5, SIGTRAP) == NOTIFY_STOP) {
1990 			return;
1991 		}
1992 		if (debugger_sstep(regs))
1993 			return;
1994 	} else if (debug_status & DBSR_IC) { 	/* Instruction complete */
1995 		regs->msr &= ~MSR_DE;
1996 
1997 		/* Disable instruction completion */
1998 		mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
1999 		/* Clear the instruction completion event */
2000 		mtspr(SPRN_DBSR, DBSR_IC);
2001 
2002 		if (kprobe_post_handler(regs))
2003 			return;
2004 
2005 		if (notify_die(DIE_SSTEP, "single_step", regs, 5,
2006 			       5, SIGTRAP) == NOTIFY_STOP) {
2007 			return;
2008 		}
2009 
2010 		if (debugger_sstep(regs))
2011 			return;
2012 
2013 		if (user_mode(regs)) {
2014 			current->thread.debug.dbcr0 &= ~DBCR0_IC;
2015 			if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0,
2016 					       current->thread.debug.dbcr1))
2017 				regs->msr |= MSR_DE;
2018 			else
2019 				/* Make sure the IDM bit is off */
2020 				current->thread.debug.dbcr0 &= ~DBCR0_IDM;
2021 		}
2022 
2023 		_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
2024 	} else
2025 		handle_debug(regs, debug_status);
2026 }
2027 NOKPROBE_SYMBOL(DebugException);
2028 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */
2029 
2030 #if !defined(CONFIG_TAU_INT)
2031 void TAUException(struct pt_regs *regs)
2032 {
2033 	printk("TAU trap at PC: %lx, MSR: %lx, vector=%lx    %s\n",
2034 	       regs->nip, regs->msr, regs->trap, print_tainted());
2035 }
2036 #endif /* CONFIG_INT_TAU */
2037 
2038 #ifdef CONFIG_ALTIVEC
2039 void altivec_assist_exception(struct pt_regs *regs)
2040 {
2041 	int err;
2042 
2043 	if (!user_mode(regs)) {
2044 		printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode"
2045 		       " at %lx\n", regs->nip);
2046 		die("Kernel VMX/Altivec assist exception", regs, SIGILL);
2047 	}
2048 
2049 	flush_altivec_to_thread(current);
2050 
2051 	PPC_WARN_EMULATED(altivec, regs);
2052 	err = emulate_altivec(regs);
2053 	if (err == 0) {
2054 		regs->nip += 4;		/* skip emulated instruction */
2055 		emulate_single_step(regs);
2056 		return;
2057 	}
2058 
2059 	if (err == -EFAULT) {
2060 		/* got an error reading the instruction */
2061 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2062 	} else {
2063 		/* didn't recognize the instruction */
2064 		/* XXX quick hack for now: set the non-Java bit in the VSCR */
2065 		printk_ratelimited(KERN_ERR "Unrecognized altivec instruction "
2066 				   "in %s at %lx\n", current->comm, regs->nip);
2067 		current->thread.vr_state.vscr.u[3] |= 0x10000;
2068 	}
2069 }
2070 #endif /* CONFIG_ALTIVEC */
2071 
2072 #ifdef CONFIG_FSL_BOOKE
2073 void CacheLockingException(struct pt_regs *regs, unsigned long address,
2074 			   unsigned long error_code)
2075 {
2076 	/* We treat cache locking instructions from the user
2077 	 * as priv ops, in the future we could try to do
2078 	 * something smarter
2079 	 */
2080 	if (error_code & (ESR_DLK|ESR_ILK))
2081 		_exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
2082 	return;
2083 }
2084 #endif /* CONFIG_FSL_BOOKE */
2085 
2086 #ifdef CONFIG_SPE
2087 void SPEFloatingPointException(struct pt_regs *regs)
2088 {
2089 	extern int do_spe_mathemu(struct pt_regs *regs);
2090 	unsigned long spefscr;
2091 	int fpexc_mode;
2092 	int code = FPE_FLTUNK;
2093 	int err;
2094 
2095 	/* We restore the interrupt state now */
2096 	if (!arch_irq_disabled_regs(regs))
2097 		local_irq_enable();
2098 
2099 	flush_spe_to_thread(current);
2100 
2101 	spefscr = current->thread.spefscr;
2102 	fpexc_mode = current->thread.fpexc_mode;
2103 
2104 	if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
2105 		code = FPE_FLTOVF;
2106 	}
2107 	else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
2108 		code = FPE_FLTUND;
2109 	}
2110 	else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
2111 		code = FPE_FLTDIV;
2112 	else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
2113 		code = FPE_FLTINV;
2114 	}
2115 	else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
2116 		code = FPE_FLTRES;
2117 
2118 	err = do_spe_mathemu(regs);
2119 	if (err == 0) {
2120 		regs->nip += 4;		/* skip emulated instruction */
2121 		emulate_single_step(regs);
2122 		return;
2123 	}
2124 
2125 	if (err == -EFAULT) {
2126 		/* got an error reading the instruction */
2127 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2128 	} else if (err == -EINVAL) {
2129 		/* didn't recognize the instruction */
2130 		printk(KERN_ERR "unrecognized spe instruction "
2131 		       "in %s at %lx\n", current->comm, regs->nip);
2132 	} else {
2133 		_exception(SIGFPE, regs, code, regs->nip);
2134 	}
2135 
2136 	return;
2137 }
2138 
2139 void SPEFloatingPointRoundException(struct pt_regs *regs)
2140 {
2141 	extern int speround_handler(struct pt_regs *regs);
2142 	int err;
2143 
2144 	/* We restore the interrupt state now */
2145 	if (!arch_irq_disabled_regs(regs))
2146 		local_irq_enable();
2147 
2148 	preempt_disable();
2149 	if (regs->msr & MSR_SPE)
2150 		giveup_spe(current);
2151 	preempt_enable();
2152 
2153 	regs->nip -= 4;
2154 	err = speround_handler(regs);
2155 	if (err == 0) {
2156 		regs->nip += 4;		/* skip emulated instruction */
2157 		emulate_single_step(regs);
2158 		return;
2159 	}
2160 
2161 	if (err == -EFAULT) {
2162 		/* got an error reading the instruction */
2163 		_exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
2164 	} else if (err == -EINVAL) {
2165 		/* didn't recognize the instruction */
2166 		printk(KERN_ERR "unrecognized spe instruction "
2167 		       "in %s at %lx\n", current->comm, regs->nip);
2168 	} else {
2169 		_exception(SIGFPE, regs, FPE_FLTUNK, regs->nip);
2170 		return;
2171 	}
2172 }
2173 #endif
2174 
2175 /*
2176  * We enter here if we get an unrecoverable exception, that is, one
2177  * that happened at a point where the RI (recoverable interrupt) bit
2178  * in the MSR is 0.  This indicates that SRR0/1 are live, and that
2179  * we therefore lost state by taking this exception.
2180  */
2181 void unrecoverable_exception(struct pt_regs *regs)
2182 {
2183 	pr_emerg("Unrecoverable exception %lx at %lx (msr=%lx)\n",
2184 		 regs->trap, regs->nip, regs->msr);
2185 	die("Unrecoverable exception", regs, SIGABRT);
2186 }
2187 NOKPROBE_SYMBOL(unrecoverable_exception);
2188 
2189 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x)
2190 /*
2191  * Default handler for a Watchdog exception,
2192  * spins until a reboot occurs
2193  */
2194 void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
2195 {
2196 	/* Generic WatchdogHandler, implement your own */
2197 	mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
2198 	return;
2199 }
2200 
2201 void WatchdogException(struct pt_regs *regs)
2202 {
2203 	printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
2204 	WatchdogHandler(regs);
2205 }
2206 #endif
2207 
2208 /*
2209  * We enter here if we discover during exception entry that we are
2210  * running in supervisor mode with a userspace value in the stack pointer.
2211  */
2212 void kernel_bad_stack(struct pt_regs *regs)
2213 {
2214 	printk(KERN_EMERG "Bad kernel stack pointer %lx at %lx\n",
2215 	       regs->gpr[1], regs->nip);
2216 	die("Bad kernel stack pointer", regs, SIGABRT);
2217 }
2218 NOKPROBE_SYMBOL(kernel_bad_stack);
2219 
2220 void __init trap_init(void)
2221 {
2222 }
2223 
2224 
2225 #ifdef CONFIG_PPC_EMULATED_STATS
2226 
2227 #define WARN_EMULATED_SETUP(type)	.type = { .name = #type }
2228 
2229 struct ppc_emulated ppc_emulated = {
2230 #ifdef CONFIG_ALTIVEC
2231 	WARN_EMULATED_SETUP(altivec),
2232 #endif
2233 	WARN_EMULATED_SETUP(dcba),
2234 	WARN_EMULATED_SETUP(dcbz),
2235 	WARN_EMULATED_SETUP(fp_pair),
2236 	WARN_EMULATED_SETUP(isel),
2237 	WARN_EMULATED_SETUP(mcrxr),
2238 	WARN_EMULATED_SETUP(mfpvr),
2239 	WARN_EMULATED_SETUP(multiple),
2240 	WARN_EMULATED_SETUP(popcntb),
2241 	WARN_EMULATED_SETUP(spe),
2242 	WARN_EMULATED_SETUP(string),
2243 	WARN_EMULATED_SETUP(sync),
2244 	WARN_EMULATED_SETUP(unaligned),
2245 #ifdef CONFIG_MATH_EMULATION
2246 	WARN_EMULATED_SETUP(math),
2247 #endif
2248 #ifdef CONFIG_VSX
2249 	WARN_EMULATED_SETUP(vsx),
2250 #endif
2251 #ifdef CONFIG_PPC64
2252 	WARN_EMULATED_SETUP(mfdscr),
2253 	WARN_EMULATED_SETUP(mtdscr),
2254 	WARN_EMULATED_SETUP(lq_stq),
2255 	WARN_EMULATED_SETUP(lxvw4x),
2256 	WARN_EMULATED_SETUP(lxvh8x),
2257 	WARN_EMULATED_SETUP(lxvd2x),
2258 	WARN_EMULATED_SETUP(lxvb16x),
2259 #endif
2260 };
2261 
2262 u32 ppc_warn_emulated;
2263 
2264 void ppc_warn_emulated_print(const char *type)
2265 {
2266 	pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm,
2267 			    type);
2268 }
2269 
2270 static int __init ppc_warn_emulated_init(void)
2271 {
2272 	struct dentry *dir, *d;
2273 	unsigned int i;
2274 	struct ppc_emulated_entry *entries = (void *)&ppc_emulated;
2275 
2276 	if (!powerpc_debugfs_root)
2277 		return -ENODEV;
2278 
2279 	dir = debugfs_create_dir("emulated_instructions",
2280 				 powerpc_debugfs_root);
2281 	if (!dir)
2282 		return -ENOMEM;
2283 
2284 	d = debugfs_create_u32("do_warn", 0644, dir,
2285 			       &ppc_warn_emulated);
2286 	if (!d)
2287 		goto fail;
2288 
2289 	for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++) {
2290 		d = debugfs_create_u32(entries[i].name, 0644, dir,
2291 				       (u32 *)&entries[i].val.counter);
2292 		if (!d)
2293 			goto fail;
2294 	}
2295 
2296 	return 0;
2297 
2298 fail:
2299 	debugfs_remove_recursive(dir);
2300 	return -ENOMEM;
2301 }
2302 
2303 device_initcall(ppc_warn_emulated_init);
2304 
2305 #endif /* CONFIG_PPC_EMULATED_STATS */
2306